Running a model on your own hardware or servers rather than accessing it through a cloud service or API.